Journals / Journal of Information Systems and Management Research / 2026 / Cilt: 8 Sayı: 1

A BERT-Based and Triage-Oriented Decision Support System for Disaster Emergency Detection from Turkish X (Twitter) Posts

Türkçe X (Twitter) Paylaşımlarından Afet Acil Durum Tespiti için BERT Temelli ve Triyaj Odaklı Bir Karar Destek Sistemi

Şura Özkan(GAZİ ÜNİVERSİTESİ)0009-0002-0539-4750
İsmail Atacak(GAZİ ÜNİVERSİTESİ)0000-0002-6357-0073
Pages
24–38
Publication Date
2026-06-30
🏛️Bibliographic metadata and full text access for this article are provided via DergiPark.
DergiPark Record ↗
✨

AI Summary & Key Takeaways

AI Insight
⚡ TL;DR(Too Long; Didn't Read)

Consequently, a decision support system was developed that can effectively evaluate disaster processes using social media data and has the potential for implementation by response teams.

📌 3 Key Takeaways

🎯
Objective & Hypothesis:Social media platforms are critical data sources that can contribute to the management of the process with the instant information flow they provide during disasters and crises.
📊
Key Findings & Data:Measurements based on the confusion matrix showed that the DDS achieved an accuracy of 88.14% and an F-score of 88.78%, demonstrating successful and stable performance for disaster emergency case detection.
💡
Conclusion & Impact:Consequently, a decision support system was developed that can effectively evaluate disaster processes using social media data and has the potential for implementation by response teams.
🔬Design:Modeling & Simulation
📍Scope:Comprehensive Sample
⚙️Variables:Sosyal Medya Analizi · BERT · Afet Yönetimi

Abstract

Social media platforms are critical data sources that can contribute to the management of the process with the instant information flow they provide during disasters and crises. In general, in cases where such a process occurs, high message density, misleading, incomplete, or noisy information flow arising from its nature are important factors that complicate the management of this process. Most studies in the literature on this topic have focused on the classification process while neglecting triage-oriented case prioritization. In this context, to overcome the related limitations, a transformer-based and triage-oriented decision support system (DSS) was developed to detect disaster emergencies from Turkish X posts. In the study, a pre-trained BERT-based deep learning (DL) model specifically tailored for the Turkish language was used to determine the probability of emergencies. The model outputs were supported by steps for identifying the type of need, extracting location information, and calculating urgency levels based on contextual clues. By combining these components, a priority score was calculated for each message for operational use, and the messages were ranked according to a triage approach. Experimental studies were conducted by applying the BERT model and proposed DSS to a dataset consisting of messages shared during the İzmir and Hatay earthquakes. Measurements based on the confusion matrix showed that the DDS achieved an accuracy of 88.14% and an F-score of 88.78%, demonstrating successful and stable performance for disaster emergency case detection. The proposed system proved effective in capturing emergency cases, particularly with a recall score of 96.56%. In the Precision@K analyses performed on the prioritization results, the DSS demonstrated superior performance over the BERT model at increasing data scales. Consequently, a decision support system was developed that can effectively evaluate disaster processes using social media data and has the potential for implementation by response teams.

Özet

Sosyal medya platformları, afet ve kriz durumlarında sahadan sağladıkları anlık bilgi akışı ile sürecin yönetimine katkı sağlayabilecek kritik veri kaynaklarındandır. Genel olarak böyle bir sürecin gerçekleştiği durumlarda onun doğası kaynaklı oluşan yüksek mesaj yoğunluğu, yanıltıcı, eksik yada gürültülü bilgi akışı bu sürecin yönetimini zorlaştıran önemli etkenler olarak karşımıza çıkar. Literatürde ilgili konu üzerinde yapılan çalışmaların çoğu sınıflandırma sürecine odaklanmış olup triaj odaklı vaka önceliklendirmesini gözardı etmektedir. Bu kapsamda bu çalışmada ilgili sınırlıkların üstesinden gelebilmek için Türkçe X paylaşımlarından afet acil durum tespiti yapan transformer temelli ve triyaj odaklı bir karar destek sistemi (KDS) geliştirilmiştir. Çalışmada, acil durum olasılığını belirlemek için Türkçe diline özgü olarak önceden eğitilmiş BERT temelli bir derin öğrenme (DÖ) modeli kullanılmıştır. Model çıktıları; ihtiyaç türü belirleme, konum bilgisi çıkarımı ve bağlamsal ipuçlarına dayalı aciliyet seviyesi hesaplama adımları ile desteklenmiştir. Bu bileşenler birleştirilerek her bir mesaj için operasyonel kullanım amacıyla bir öncelik puanı hesaplanarak, mesajlar triyaj yaklaşımına uygun olarak sıralanmıştır. Deneysel çalışmalar, BERT modeli ve önerilen KDS’ nin İzmir ve Hatay depremleri sırasında paylaşılan mesajından oluşan veri kümesine uygulanması ile gerçekleştirilmiştir. Karışıklık matrisi temelindeki ölçümler, KDS’ nin elde ettiği % 88.14’lük doğruluk ve % 88.78’lik f-ölçütü skorları ile afet acil durum tespiti için başarılı ve istikrarlı bir performans ortaya koyduğunu göstermiştir. Önerilen sistem, özellikle duyarlılık metriğinde elde ettiği %96.56’ lık skor ile acil durum vakalarını yakalamada ne kadar etkili olduğunu ispatlamıştır. Önceliklendirme sonuçları üzerinden gerçekleştirilen Precision@K analizlerinde KDS’ nin artan veri ölçeğinde BERT modelinin üstünde bir performansa imza attığı gözlenmiştir. Sonuç olarak, önerilen sistem ile sosyal medya verileri üzerinden afet sürecini etkin biçimde değerlendirebilen ve müdahale ekipleri için uygulanabilme potansiyeline sahip bir karar destek sistemi ortaya konmuştur.

🔖

Cite This Article

Özkan, Ş., Atacak, İ. (2026). A BERT-Based and Triage-Oriented Decision Support System for Disaster Emergency Detection from Turkish X (Twitter) Posts. Journal of Information Systems and Management Research, Cilt: 8 Sayı: 1, 24–38. https://doi.org/10.59940/jismar.1907364

📄 Full Text PDF Viewer

📄

PDF görüntüleyici hazırlanıyor...

References (30)

  1. 1] O. Dodanlı, “Afetlerde Dezenformasyonla Mücadele: Zorluklar ve Stratejiler,” İdarecinin Sesi Dergisi, 2024, Accessed: Jan. 03, 2026. [Online]. Available: https://www.academia.edu/126608060/Afetlerde_Dezenformasyonla_M%C3%BCcadele_Zorluklar_ve_Stratejiler
  2. A. Keleş, “Afet Anında Doğru Bilgiye Ulaşım,” Medium, Jan. 2024, Accessed: Jan. 03, 2026. [Online]. Available: https://alpkeles99.medium.com/afet-anında-doğru-bilgiye-ulaşım-e3646bd5a3bb
  3. S. K. Abid, R. Roosli, U. Nazir, and N. S. Kamarudin, “AI-enhanced crowdsourcing for disaster management: strengthening community resilience through social media,” International Journal of Emergency Medicine 2025 18:1, vol. 18, no. 1, pp. 201-, Oct. 2025, doi: 10.1186/s12245-025-01009-9.
  4. C. He and D. Hu, “Social Media Analytics for Disaster Response: Classification and Geospatial Visualization Framework,” Applied Sciences 2025, Vol. 15, Page 4330, vol. 15, no. 8, p. 4330, Apr. 2025, doi: 10.3390/app15084330.
  5. S. Vieweg, A. L. Hughes, K. Starbird, and L. Palen, “Microblogging during two natural hazards events: What twitter may contribute to situational awareness,” Conference on Human Factors in Computing Systems - Proceedings, vol. 2, pp. 1079–1088, 2010, doi: 10.1145/1753326.1753486.
  6. S. N. Zisad, N. M. I. Chowdhury, and R. Hasan, “Comparative Analysis of Transformer Models in Disaster Tweet Classification for Public Safety,” Sep. 2025, Accessed: Mar. 01, 2026. [Online]. Available: https://arxiv.org/pdf/2509.04650v1
  7. S. Aslan and M. Yildirim, “A novel attention-based deep learning model for improving sentiment classification after the case of the 2023 Kahramanmaras/Turkey earthquake on Twitter,” PeerJ Comput. Sci., vol. 11, p. e2881, May 2025, doi: 10.7717/peerj-cs.2881.
  8. M. Söylemez and A. Öztürk, “Detection of Urgent Messages Shared on Twitter during an Earthquake using the Deep Learning Method,” Journal of Data Applications, vol. 0, no. 3, pp. 63–84, Jan. 2025, doi: 10.26650/joda.1606403.
  9. K. S. Gill, V. Anand, D. Upadhyay, and S. Dangi, “Classification of Tweets using a Machine Learning and Natural Language Processing Algorithm for Disaster Prediction,” 2024 3rd International Conference for Innovation in Technology, INOCON 2024, 2024, doi: 10.1109/INOCON60754.2024.10512145.
  10. A. Devaraj, D. Murthy, and A. Dontula, “Machine-learning methods for identifying social media-based requests for urgent help during hurricanes,” International Journal of Disaster Risk Reduction, vol. 51, no. 2, p. 101757, Dec. 2020, doi: 10.1016/j.ijdrr.2020.101757.
  11. S. Behl, A. Rao, S. Aggarwal, S. Chadha, and H. S. Pannu, “Twitter for disaster relief through sentiment analysis for COVID-19 and natural hazard crises,” International Journal of Disaster Risk Reduction, vol. 55, no. 1, p. 102101, Mar. 2021, doi: 10.1016/j.ijdrr.2021.102101.
  12. F. Yao and Y. Wang, “Domain-specific sentiment analysis for tweets during hurricanes (DSSA-H): A domain-adversarial neural-network-based approach,” Comput. Environ. Urban Syst., vol. 83, no. 03, p. 101522, Sep. 2020, doi: 10.1016/j.compenvurbsys.2020.101522.
  13. B. J. Detera, A. Kodaka, N. Kohtake, A. Nishino, and K. Onda, “An English-Japanese Twitter-Based Analysis of Disaster Sentiment during Typhoons and Earthquakes,” ISSE 2021 - 7th IEEE International Symposium on Systems Engineering, Proceedings, Sep. 2021, doi: 10.1109/ISSE51541.2021.9582473.
  14. V. Balakrishnan, Z. Shi, C. L. Law, R. Lim, L. L. Teh, Y. Fan and J. Periasamy, “A Comprehensive Analysis of Transformer-Deep Neural Network Models in Twitter Disaster Detection,” Mathematics, vol. 10, no. 24, p. 4664, Dec. 2022, doi: 10.3390/math10244664.
  15. P. Duraisamy and Y. Natarajan, “Twitter Disaster Prediction Using Different Deep Learning Models,” SN Computer Science 2024 5:1, vol. 5, no. 1, pp. 179-, Jan. 2024, doi: 10.1007/s42979-023-02520-7.
  16. D. S. Krishna, G. Srinivas, and P. V. G. D. Prasad Reddy, “A Deep Parallel Hybrid Fusion Model for disaster tweet classification on Twitter data,” Decision Analytics Journal, vol. 11, p. 100453, Jun. 2024, doi: 10.1016/j.dajour.2024.100453.
  17. D. S. Krishna, G. Srinivas, and P. V. G. D. Prasad Reddy, “Disaster tweet classification: A majority voting approach using machine learning algorithms,” Intelligent Decision Technologies, vol. 17, no. 2, pp. 343–355, 2023, doi: 10.3233/IDT-220310.
  18. K. Yin, B. Li, C. Liu, A. Mostafavi, and X. Hu, “CrisisSense-LLM: Instruction Fine-Tuned Large Language Model for Multi-label Social Media Text Classification in Disaster Informatics,” ArXiv, p. arXiv:2406.15477, Jun. 2024, Accessed: Mar. 02, 2026. [Online]. Available: https://arxiv.org/pdf/2406.15477
  19. H. Jinzhen, K. Jisung, Y. J. Soo, and Y. H. Sik, “A Lightweight LLM Framework for Disaster Humanitarian Information Classification,” Jan. 2026, Accessed: Mar. 02, 2026. [Online]. Available: http://arxiv.org/abs/2602.12284
  20. J. Pereira, R. Lotufo, and R. Nogueira, “Large Language Models in Summarizing Social Media for Emergency Management,” in Proceedings of the 32nd Text Retrieval Conference (TREC 2023), I.; E. A. Soboroff, Ed., 2023. doi: 10.26650/JODA.1606403.
  21. Y. Wang, C. Wang, X. Zhang, and L. Zeng, “Towards Intelligent Emergency Management: A Scenario–Learning–Decision Framework Enabled by Large Language Models,” Mathematics 2025, Vol. 13, Page 3463, vol. 13, no. 21, p. 3463, Oct. 2025, doi: 10.3390/math13213463.
  22. A. B. Can, “Earthquake Turkish Tweets Partially Annotated, Dataset,” Kaggle. Accessed: Oct. 16, 2025. [Online]. Available: https://www.kaggle.com/datasets/aliburakcan/earthquake-turkish-tweets-partially-annotated
  23. dbmdz, “bert-base-turkish-cased,” Hugging Face. Accessed: Oct. 01, 2025. [Online]. Available: https://huggingface.co/dbmdz/bert-base-turkish-cased
  24. İ. Atacak and Ö. A. Şencan, “Mamdani ve Sugeno Tip Bulanık Çıkarım Sistemleri ile Sosyal Medya Haber Popülerliğinin Tahmini,” International Journal of Engineering Research and Development, vol. 14, no. 3, pp. 303–320, Dec. 2022, doi: 10.29137/umagd.1169623.
  25. İ. Atacak, “Kalp Yetmezliği Tahmininin Kategorik Olarak Farklı Tip Makine Öğrenmesi Yöntemleri ile Uygulanmasına Yönelik Bir Değerlendirme Çalışması,” EMO Bilimsel Dergi, vol. 14, no. 1, 2024.
  26. V. Jayaswal, “Performance Metrics: Confusion matrix, Precision, Recall, and F1 Score,” Towards Data Science. Accessed: Jan. 12, 2026. [Online]. Available: https://towardsdatascience.com/performance-metrics-confusion-matrix-precision-recall-and-f1-score-a8fe076a2262/
  27. “Understanding the Confusion Matrix in Machine Learning,” GeeksforGeeks. Accessed: Jan. 23, 2026. [Online]. Available: https://www.geeksforgeeks.org/machine-learning/confusion-matrix-machine-learning/
  28. K. Pullakandam, “Understanding Precision, Recall, and F-Score at K in Recommender Systems,” Medium. Accessed: Jan. 23, 2026. [Online]. Available: https://krishnapullak.medium.com/understanding-precision-recall-and-f-score-at-k-in-recommender-systems-7146a0dce68e
  29. Wikipedia contributors, “Evaluation measures (information retrieval),” Wikipedia: The Free Encyclopedia. Accessed: Jan. 22, 2026. [Online]. Available: https://en.wikipedia.org/wiki/Evaluation_measures_(information_retrieval).
  30. J. Pereira, R. Lotufo, and R. Nogueira, “Large Language Models in Summarizing Social Media for Emergency Management,” in Proceedings of the 32nd Text Retrieval Conference (TREC 2023), I.; E. A. Soboroff, Ed., 2023. doi: 10.26650/JODA.1606403.